Perspectives for Direct Interpretability in Multi-Agent Deep Reinforcement Learning